Ornith-1.0: Self-Scaffolding LLMs for Agentic Coding

DeepReinforce has released Ornith-1.0, an open-weight model licensed under MIT that achieves state-of-the-art performance among open-source models of comparable size on coding benchmarks. The model is built upon pretrained Gemma 4 and Qwen 3.5 foundations and includes variants with 9B Dense, 31B Dense, 35B MoE, and 397B MoE parameter counts.

• Ornith-1.0 utilizes self-scaffolding techniques for agentic coding tasks.
• It is compatible with Apache 2.0 licensed underlying models (Gemma 4 and Qwen 3.5).
• The 35B MoE variant runs efficiently in local environments like LM Studio, handling complex tool calls proficiently.

The release provides a high-performing option for local LLM deployment, demonstrating strong capabilities in navigating codebases and executing multi-step agent harnesses.